Re: Re: MemoryView questions

On 8 August 2012 16:24, Dave Hirschfeld <dave.hirschfeld <at> gmail.com> wrote:
> Bradley Froehle <brad.froehle <at> gmail.com> writes:
>
>>
>>
>> Hi Folks,
>>
>> I've got four questions relating to memoryviews in Cython 0.17.beta1, which
> hopefully you can help me with.
>> 1. Can typed memory views accept multiple dimensions?
>> I've noticed that memory views do have an ndim attribute, however defining a
> function as:
>>     def f(double[:] x):
>>         pass
>> only works if x is one-dimensional.  Can I somehow allow x to have 1, 2, or 3
> dimensions?  I'm fine adding an `assert 1 <= x.ndim <= 3` statement.
>>
>> 2. If I have a multidimensional array, why do I have to call `is_f_contig` to
> see if it is f-contiguous?  It seems like that should be a property, much like
> `ndim`.
>>
>>     def f(double[:, :] x):
>>
>>         print x.is_f_contig()
>>
>> 3. Supposing that I know in advance that my memory is contiguous, how can I
> get access to the underlying data pointer of the memoryview?
>>
>> Certainly I could do:
>>     def f(double[::1, :] x):
>>         cfunction(&x[0,0])
>> ... but what if I don't know the dimensionality of x?  (See Q 1).
>>
>>
>> Perhaps there should be a x.data attribute?
>>
>> 4. Can I not have inline memoryview functions in pxd files?
>>
>> Consider:
>>
>> ### lib.pxd ###
>> cdef inline void f(double[:,:] x):
>>     pass
>> cdef inline f1(x):
>>     f(x)
>>
>> ### tst.pyx ###
>> cimport lib
>>
>> Then compiling tst.pyx fails:
>>
>>
>> $ python setup.py build_ext --inplace
>> running build_ext
>> building 'tst' extension
>> creating build
>> creating build/temp.linux-x86_64-2.7
>> gcc -pthread -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-
> prototypes -fPIC -I/usr/include/python2.7 -c tst.c -o build/temp.linux-x86_64-
> 2.7/tst.o
>> tst.c: In function ‘__pyx_f_3lib_f1’:
>> tst.c:1232:3: warning: implicit declaration of function
> ‘__Pyx_PyObject_to_MemoryviewSlice_dsds_double’ [-Wimplicit-function-
> declaration]
>> tst.c:1232:13: error: incompatible types when assigning to type
> ‘__Pyx_memviewslice’ from type ‘int’
>> tst.c: At top level:
>> tst.c:13446:12: warning: ‘__Pyx_ValidateAndInit_memviewslice’ defined but not
> used [-Wunused-function]
>> error: command 'gcc' failed with exit status 1
>>
>>
>> Thanks!
>> -Brad
>>
>
> Not sure about the answers to your questions in general, but I thought you
> might be able to cheat by using fused types. Unfortunately this seems not
> to work:
>
> In [32]: %%cython
>     ...: cimport cython
>     ...:
>     ...: from cpython cimport bool
>     ...:
>     ...: import numpy as np
>     ...: cimport numpy as np
>     ...:
>     ...: ctypedef np.float64_t float64_t
>     ...:
>     ...: ctypedef fused myarray:
>     ...:     float64_t[:,:]
>     ...:     float64_t[:,:,:]
>     ...: #
>     ...:
>     ...: cpdef myarray nop(myarray array):
>     ...:     return array
>     ...:
>
> In [33]: nop(rand(2))
> ---------------------------------------------------------------------------
> TypeError                                 Traceback (most recent call last)
> <ipython-input-33-bbba0b03d22e> in <module>()
> ----> 1 nop(rand(2))
>
> f.pyd in f.__pyx_fused_cpdef (f.c:2503)()
>
> TypeError: No matching signature found
>
> In [34]: nop(rand(2,2))
> ---------------------------------------------------------------------------
> TypeError                                 Traceback (most recent call last)
> <ipython-input-34-fa44702cf241> in <module>()
> ----> 1 nop(rand(2,2))
>
> f.pyd in f.__pyx_fused_cpdef (Cf.c:2503)()
>
> TypeError: No matching signature found
>
> In [35]: nop(rand(2,2,2))
> ---------------------------------------------------------------------------
> TypeError                                 Traceback (most recent call last)
> <ipython-input-35-c6b24e4d9bf8> in <module>()
> ----> 1 nop(rand(2,2,2))
>
> f.pyd in f.__pyx_fused_cpdef f.c:2503)()
>
> TypeError: No matching signature found
>
> In [36]:
>
>
> Is there any reason why this shouldn't work or is this a bug?
>
> -Dave
>
>

Thanks, that's a bug. It seems the delimiter ', ' was wrong, since it
generates signature strings like 'double[:, :]'. It also didn't
dispatch on ndim, so that's fixed as well. There's a fix here
https://github.com/markflorisson88/cython/commit/f96a5622c68776d4a206f36d1f701c763bad7caf
which will be pushed for 0.17 if it doesn't break anything.